📊 Predicting Insurance Enrollment

A machine learning pipeline to analyze employee data and predict voluntary insurance enrollment.

🧾 Problem Statement

We aim to predict whether an employee will opt-in to a voluntary insurance plan using demographic and employment data. The dataset contains 10,000 synthetic records.

import pandas as pd
# Load dataset
df = pd.read_csv("employee_data.csv")
df.head()

	employee_id	age	gender	marital_status	salary	employment_type	region	has_dependents	tenure_years	enrolled
0	10001	60	Female	Single	55122.97	Part-time	West	No	1.5	0
1	10002	50	Female	Single	89549.66	Full-time	West	Yes	12.8	1
2	10003	36	Male	Divorced	74145.66	Part-time	Midwest	No	3.8	0
3	10004	64	Female	Married	53877.83	Full-time	Northeast	No	3.3	0
4	10005	29	Male	Single	63404.63	Contract	Midwest	Yes	10.0	0

🔍 Exploratory Data Analysis

Analyzing patterns in enrollment by different groups.

df.describe(include="all")

	employee_id	age	gender	marital_status	salary	employment_type	region	has_dependents	tenure_years	enrolled
count	10000.00000	10000.0000	10000	10000	10000.000000	10000	10000	10000	10000.000000	10000.000000
unique	NaN	NaN	3	4	NaN	3	4	2	NaN	NaN
top	NaN	NaN	Male	Married	NaN	Full-time	West	Yes	NaN	NaN
freq	NaN	NaN	4815	4589	NaN	7041	2582	5993	NaN	NaN
mean	15000.50000	43.0020	NaN	NaN	65032.967907	NaN	NaN	NaN	3.967720	0.617400
std	2886.89568	12.2858	NaN	NaN	14923.958446	NaN	NaN	NaN	3.895488	0.486046
min	10001.00000	22.0000	NaN	NaN	2207.790000	NaN	NaN	NaN	0.000000	0.000000
25%	12500.75000	33.0000	NaN	NaN	54714.342500	NaN	NaN	NaN	1.200000	0.000000
50%	15000.50000	43.0000	NaN	NaN	65056.050000	NaN	NaN	NaN	2.800000	1.000000
75%	17500.25000	54.0000	NaN	NaN	75053.687500	NaN	NaN	NaN	5.600000	1.000000
max	20000.00000	64.0000	NaN	NaN	120312.000000	NaN	NaN	NaN	36.000000	1.000000

df.isna().sum()
# non nan values

employee_id        0
age                0
gender             0
marital_status     0
salary             0
employment_type    0
region             0
has_dependents     0
tenure_years       0
enrolled           0
dtype: int64

import plotly.express as px



# Create bar charts using plotly.express
fig_gender = px.bar(
    df.groupby('gender')['enrolled'].mean().reset_index(),
    x='gender', y='enrolled',
    title='Enrollment Rate by Gender',
    labels={'enrolled': 'Enrollment Rate'}
)

fig_marital = px.bar(
    df.groupby('marital_status')['enrolled'].mean().reset_index(),
    x='marital_status', y='enrolled',
    title='Enrollment Rate by Marital Status',
    labels={'enrolled': 'Enrollment Rate'}
)

fig_employment = px.bar(
    df.groupby('employment_type')['enrolled'].mean().reset_index(),
    x='employment_type', y='enrolled',
    title='Enrollment Rate by Employment Type',
    labels={'enrolled': 'Enrollment Rate'}
)

fig_region = px.bar(
    df.groupby('region')['enrolled'].mean().reset_index(),
    x='region', y='enrolled',
    title='Enrollment Rate by Region',
    labels={'enrolled': 'Enrollment Rate'}
)

fig_dependents = px.bar(
    df.groupby('has_dependents')['enrolled'].mean().reset_index(),
    x='has_dependents', y='enrolled',
    title='Enrollment Rate by Dependents',
    labels={'enrolled': 'Enrollment Rate'}
)

# Show the figures
fig_gender.show()
fig_marital.show()
fig_employment.show()
fig_region.show()
fig_dependents.show()

df.corr(numeric_only=True)

	employee_id	age	salary	tenure_years	enrolled
employee_id	1.000000	-0.006100	0.005034	0.001356	0.015004
age	-0.006100	1.000000	0.003872	0.002220	0.268749
salary	0.005034	0.003872	1.000000	-0.003574	0.366403
tenure_years	0.001356	0.002220	-0.003574	1.000000	-0.007480
enrolled	0.015004	0.268749	0.366403	-0.007480	1.000000

fig_age = px.box(df, x='enrolled', y='age', points='all', title="Age Distribution by Enrollment")
fig_salary = px.box(df, x='enrolled', y='salary', points='all', title="Salary Distribution by Enrollment")
fig_tenure = px.box(df, x='enrolled', y='tenure_years', points='all', title="Tenure Distribution by Enrollment")

# Histograms to see distributions
fig_age_hist = px.histogram(df, x='age', color='enrolled', barmode='overlay', nbins=30, title="Age Histogram by Enrollment")
fig_salary_hist = px.histogram(df, x='salary', color='enrolled', barmode='overlay', nbins=30, title="Salary Histogram by Enrollment")
fig_tenure_hist = px.histogram(df, x='tenure_years', color='enrolled', barmode='overlay', nbins=30, title="Tenure Histogram by Enrollment")

# Binned average enrollment line plots
df['age_bin'] = pd.cut(df['age'], bins=10)
df['salary_bin'] = pd.cut(df['salary'], bins=10)
df['tenure_bin'] = pd.cut(df['tenure_years'], bins=10)

age_trend = df.groupby('age_bin')['enrolled'].mean().reset_index()
salary_trend = df.groupby('salary_bin')['enrolled'].mean().reset_index()
tenure_trend = df.groupby('tenure_bin')['enrolled'].mean().reset_index()


# Show all plots
fig_age.show()
fig_salary.show()
fig_tenure.show()

fig_age_hist.show()
fig_salary_hist.show()
fig_tenure_hist.show()

🤖 Modeling

Using Logistic Regression and Random Forest to predict enrollment.

from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report, roc_auc_score

X = df.drop(columns=["employee_id", "enrolled"])
y = df["enrolled"]
X_train, X_test, y_train, y_test = train_test_split(X, y, stratify=y, test_size=0.2, random_state=42)

categorical = ["gender", "marital_status", "employment_type", "region", "has_dependents"]
numeric = ["age", "salary", "tenure_years"]
preprocessor = ColumnTransformer([
    ("num", StandardScaler(), numeric),
    ("cat", OneHotEncoder(drop='first'), categorical)
])

logreg_pipeline = Pipeline([
    ("preprocessor", preprocessor),
    ("model", LogisticRegression(max_iter=1000))
])
logreg_pipeline.fit(X_train, y_train)
logreg_pred = logreg_pipeline.predict(X_test)
logreg_prob = logreg_pipeline.predict_proba(X_test)[:, 1]

print("Logistic Regression Report:")
print(classification_report(y_test, logreg_pred))
print("ROC AUC:", roc_auc_score(y_test, logreg_prob))

rf_pipeline = Pipeline([
    ("preprocessor", preprocessor),
    ("model", RandomForestClassifier(n_estimators=100, random_state=42))
])
rf_pipeline.fit(X_train, y_train)
rf_pred = rf_pipeline.predict(X_test)
rf_prob = rf_pipeline.predict_proba(X_test)[:, 1]

print("Random Forest Report:")
print(classification_report(y_test, rf_pred))
print("ROC AUC:", roc_auc_score(y_test, rf_prob))

Logistic Regression Report:
              precision    recall  f1-score   support

           0       0.87      0.85      0.86       765
           1       0.91      0.92      0.92      1235

    accuracy                           0.90      2000
   macro avg       0.89      0.89      0.89      2000
weighted avg       0.90      0.90      0.90      2000

ROC AUC: 0.9704935037442778
Random Forest Report:
              precision    recall  f1-score   support

           0       1.00      1.00      1.00       765
           1       1.00      1.00      1.00      1235

    accuracy                           1.00      2000
   macro avg       1.00      1.00      1.00      2000
weighted avg       1.00      1.00      1.00      2000

ROC AUC: 1.0

📌 Feature Importance

Understanding what drives enrollment decisions using Random Forest.

import numpy as np

ohe_features = rf_pipeline.named_steps["preprocessor"].transformers_[1][1].get_feature_names_out(categorical)
feature_names = numeric + list(ohe_features)
importances = rf_pipeline.named_steps["model"].feature_importances_
indices = np.argsort(importances)[::-1]

for i in indices[:10]:
    print(f"{feature_names[i]}: {importances[i]:.4f}")

salary: 0.2774
has_dependents_Yes: 0.2558
age: 0.1946
employment_type_Full-time: 0.1747
employment_type_Part-time: 0.0558
tenure_years: 0.0253
gender_Male: 0.0030
region_Northeast: 0.0022
marital_status_Married: 0.0022
region_West: 0.0022

import plotly.graph_objects as go
import numpy as np
import pandas as pd

# Extract feature names after preprocessing
ohe = logreg_pipeline.named_steps["preprocessor"].transformers_[1][1]
ohe_features = ohe.get_feature_names_out(categorical)
feature_names = numeric + list(ohe_features)

# Get coefficients from logistic regression
coefficients = logreg_pipeline.named_steps["model"].coef_[0]

# Create a DataFrame
coef_df = pd.DataFrame({
    "Feature": feature_names,
    "Coefficient": coefficients
}).sort_values("Coefficient")

# Assign colors based on coefficient sign
colors = ["red" if val < 0 else "green" for val in coef_df["Coefficient"]]

# Plot using Plotly
fig = go.Figure(go.Bar(
    x=coef_df["Coefficient"],
    y=coef_df["Feature"],
    orientation='h',
    marker=dict(color=colors)
))

fig.update_layout(
    title="Logistic Regression Feature Importance (Coefficients)",
    xaxis_title="Coefficient Value",
    yaxis_title="Feature",
    showlegend=False
)

🔧 Hyperparameter Tuning

Improve model performance via grid search.

✅ Summary

• Random Forest outperformed Logistic Regression.
• Most influential features: Employment Type, Dependents, Age,Salar
• Strong signals from family-related,age and financial attributes.

Next Steps

• Maybe check for overfitting in the solution as I am getting perfect score
• try to enhance logit model for better perfromance or hyperparamer tuning
• Try boosting models like XGBoost or LightGBM